[pymvpa] SVM weights - positive and negative values
robbenson18 at gmail.com
Tue Apr 2 14:24:19 UTC 2013
basically I look at classifier estimates attribute (clf.ca.estimates) and
at predictions (clf.ca.predictions), since the prediction is based on svm
decision function sign, you could derive the class label (predictions) from
the decision function sign (estimates).
For the interpretation you can look at a discussion on this forum, in which
I asked about weights signifiance:
On 31 March 2013 22:11, Ricky Savjani <rsavjani at gmail.com> wrote:
> Hi PyMVPA team,
> I had a quick question (or perhaps just pt of confusion) on the output of
> SVM weights using a sensitivity_analyzer. In particular, I think that the
> sign (positive vs negative) of the svm weights reflects the weighting for
> the particular target (say 'class1' vs "class2"). I think positive values
> would be associated with class1 and negative values to be associated with
> class2. But how can you be sure which target corresponds to positive values
> in the SVM weights? Is it simply alphabetically, meaning class1 would take
> on positive values and class2 negative values? Or is this entirely the
> incorrect interpretation of the SVM weights that are output by the
> sensitivity analyzer?
> Thanks very much for the help.
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Pkg-ExpPsy-PyMVPA